Fix flaky check-sources and parity gate PHP-engine abort

.github/workflows/blink_parity_mautic.yaml

@@ -72,7 +72,10 @@ jobs:
                           common: true,
                           standaloneLine: true,
                           cleanCode: true,
-                      );
+                      )
+                      // single process: parallel workers overflow the NDJSON buffer on
+                      // large mautic files and abort the PHP run (blink ignores this)
+                      ->withoutParallel();
                   PHP
                   cp /tmp/mautic/ecs.php /tmp/mautic/ecs-copy.php
 
@@ -92,7 +95,7 @@ jobs:
             - name: Compare the two trees
               run: |
                   # ratchet gate: fail once the differing share crosses this; lower it as parity improves
-                  MAX_DIFF_PERCENT=1.7
+                  MAX_DIFF_PERCENT=1.6
                   total=0
                   differ=0
                   differing_files=""

blink/.gitignore

@@ -0,0 +1,2 @@
+/blink
+/check-sources

blink/cmd/check-sources/main.go

@@ -1,6 +1,7 @@
 // Command check-sources verifies that every registered fixer links to a
 // reachable PHP-CS-Fixer source file on GitHub. It is run in CI so a renamed or
-// removed upstream rule fails the build.
+// removed upstream rule fails the build. GitHub rate-limits bursts of HEAD
+// requests with 503/429, so each URL is retried with backoff before failing.
 package main
 
 import (
@@ -12,34 +13,58 @@ import (
 	"blink/internal/fixer/rules"
 )
 
-func main() {
-	client := &http.Client{Timeout: 20 * time.Second}
-	failed := false
-
-	for _, f := range rules.All() {
-		url := f.SourceURL()
+const maxAttempts = 4
 
+// checkURL returns the final status code (or an error) for url, retrying on a
+// 5xx or 429 that GitHub returns when throttling a burst of requests.
+func checkURL(client *http.Client, url string) (int, error) {
+	var lastStatus int
+	var lastErr error
+	for attempt := 1; attempt <= maxAttempts; attempt++ {
 		req, err := http.NewRequest(http.MethodHead, url, nil)
 		if err != nil {
-			fmt.Printf("FAIL %s: %v\n", f.Name(), err)
-			failed = true
-			continue
+			return 0, err
 		}
-
+		req.Header.Set("User-Agent", "blink-check-sources")
 		resp, err := client.Do(req)
 		if err != nil {
-			fmt.Printf("FAIL %s: %v\n", f.Name(), err)
-			failed = true
-			continue
+			lastErr = err
+		} else {
+			_ = resp.Body.Close()
+			lastStatus, lastErr = resp.StatusCode, nil
+			if resp.StatusCode == http.StatusOK {
+				return resp.StatusCode, nil
+			}
+			// a 4xx other than 429 is a genuine missing/renamed source, not throttling
+			if resp.StatusCode != http.StatusTooManyRequests && resp.StatusCode < 500 {
+				return resp.StatusCode, nil
+			}
 		}
-		_ = resp.Body.Close()
+		if attempt < maxAttempts {
+			time.Sleep(time.Duration(attempt) * 2 * time.Second)
+		}
+	}
+	return lastStatus, lastErr
+}
 
-		if resp.StatusCode != http.StatusOK {
-			fmt.Printf("FAIL %s: %s -> %d\n", f.Name(), url, resp.StatusCode)
+func main() {
+	client := &http.Client{Timeout: 20 * time.Second}
+	failed := false
+
+	for _, f := range rules.All() {
+		url := f.SourceURL()
+		status, err := checkURL(client, url)
+		switch {
+		case err != nil:
+			fmt.Printf("FAIL %s: %v\n", f.Name(), err)
+			failed = true
+		case status != http.StatusOK:
+			fmt.Printf("FAIL %s: %s -> %d\n", f.Name(), url, status)
 			failed = true
-			continue
+		default:
+			fmt.Printf("OK   %s\n", url)
 		}
-		fmt.Printf("OK   %s\n", url)
+		time.Sleep(150 * time.Millisecond) // throttle so GitHub does not rate-limit the burst
 	}
 
 	if failed {